Android On-Device Unified STT Framework

Whisper.cpp, Vosk, and Sherpa-ONNX unified with Speaker Diarization and 0 external ML dependencies on Termux.

🎙️ Live On-Device Speech Demo Available! Listen to authentic continuous speech transcribed with Whisper Base and explore live SRT subtitle generation.

▶ Open Live Showcase

1-Line Quick Installation (Choose Language)

Select your runtime and run the 1-line installation command in Termux:

🐍 Python Edition (PyPI):

pip install termux-stt && termux-stt-install

☕ Node.js / TypeScript Edition (npm):

npm install termux-stt && npx termux-stt install

3-Line Code Usage

from termux_stt import create_engine

engine = create_engine("whisper", model="base", lang="ko")
result = engine.transcribe("meeting.wav")
print(result.text)

const { createEngine } = require("termux-stt");

const engine = createEngine("whisper", { model: "base", lang: "ko" });
const result = await engine.transcribe("meeting.wav");
console.log(result.text);

from termux_stt import create_engine

# Hybrid Pipeline: Vosk X-Vector + Whisper STT
engine = create_engine("hybrid", lang="ko", num_speakers=2)
result = engine.diarize("interview.wav")

for seg in result.segments:
    print(f"[{seg.speaker}] ({seg.start:.1f}s-{seg.end:.1f}s) {seg.text}")

Engine Comparison Matrix

AMEVA Foundation — Mobile AI & Automation Ecosystem

AMEVA Foundation Initiative: 100% on-device local AI for everyone with zero cloud subscriptions and zero data leakage.

Engine	Model	Peak RAM	RTF	Accuracy (KO)	Diarization	Mobile Rating
whisper.cpp	ggml-tiny	~150 MB	0.80	85%	❌ None	⭐⭐⭐⭐⭐
whisper.cpp	ggml-base	~250 MB	1.20	88%	❌ None	⭐⭐⭐⭐
Vosk	small-ko	~100 MB	0.25	78%	✅ X-Vector 128d	⭐⭐⭐
Sherpa-ONNX	Zipformer	~300 MB	0.42	86%	✅ CAM++	⭐⭐⭐⭐
Hybrid (Vosk+Whisper)	small-ko + base	~350 MB	1.45	92%+	✅ Built-in (K-Means)	⭐⭐⭐⭐⭐ (Recommended)

Project	Packages & Links	Core On-Device Capability
🎙️ termux-stt	PyPI • npm	Unified STT + Pure-Python 128d X-Vector Speaker Diarization
🎨 termux-diffusion	PyPI • npm	On-Device Stable Diffusion Image Generation (ARM NEON)
🌐 termux-playwright	PyPI • npm	Native Headless Chromium Automation & Scraping (Zero PRoot)
🧠 termux-train	GitHub	Pure C++ Autograd Backprop Neural Training & LoRA on Android
🖥️ AMEVA Workstation	Web App	100% On-Device WebGPU Document Intelligence Workspace
⚡ AMEVA-Forge	Live Studio	Real-Time WebGPU 3D Neural Studio & Visualization Engine